Pengembangan BPMN Sketch Recognition dengan Transformer-Based OCR dan Modul Postprocessing Berbasis Aturan BPMN 2.0

Rizky, Naufal Khairul (2026) Pengembangan BPMN Sketch Recognition dengan Transformer-Based OCR dan Modul Postprocessing Berbasis Aturan BPMN 2.0. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025221127-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (23MB) | Request a copy

Abstract

Pemodelan proses bisnis menggunakan Business Process Model and Notation (BPMN) merupakan standar dokumentasi utama dalam rekayasa perangkat lunak, yang umumnya diawali dengan sketsa tangan karena cepat dan kolaboratif. Namun, digitalisasinya ke format XML yang valid masih dilakukan manual dan tidak efisien. Penelitian terdahulu Sketch2Process menawarkan solusi BPMN Sketch Recognition, tetapi memiliki ketergantungan pada layanan Optical Character Recognition (OCR) berbayar serta rentan menghasilkan luaran yang melanggar spesifikasi BPMN 2.0 dan tidak memenuhi standar validitas struktural (block-structured). Penelitian ini mengintegrasikan model Transformer-based OCR (TrOCR) untuk menghilangkan ketergantungan Application Programming Interface (API) eksternal, sekaligus memperkenalkan modul postprocessing berbasis aturan BPMN 2.0 dan prinsip block-structured workflow untuk memperbaiki struktur akhir. Model TrOCR mencapai Character Error Rate (CER) 7,11%, mengungguli baseline acuan (8,80%). Pengujian sistem keseluruhan pada citra uji mencapai Overall F1-Score 86,0%, dengan rincian Shape F1-Score 91,5%, Edge F1-Score 80,6%, dan Label F1-Score 84,8%. Meskipun capaian per kategori di bawah penelitian acuan, modul postprocessing berhasil mengeksekusi perbaikan struktural dengan Fix Rate 100% pada sebelas kasus uji, serta False Positive Rate 0% pada gambar yang telah valid tanpa kesalahan. Analisis trade-off multi-kriteria menggunakan metode Complex Proportional Assessment (COPRAS) pada empat kriteria menunjukkan bahwa sistem usulan meraih tingkat utilitas yang lebih tinggi dibandingkan penelitian acuan pada seluruh skenario sensitivitas bobot yang diuji.
=================================================================================================================================
Business process modeling using Business Process Model and Notation (BPMN) is a primary documentation standard in software engineering, commonly initiated through hand-drawn sketches for their speed and collaborative nature. However, digitizing these sketches into valid XML remains largely manual and inefficient. A prior study, Sketch2Process, proposed a BPMN Sketch Recognition solution but depended on a paid Optical Character Recognition (OCR) service and was prone to producing output that violated the BPMN 2.0 specification and failed to meet block-structured validity standards. This research integrates a Transformer-based OCR (TrOCR) model to eliminate dependency on external Application Programming Interfaces (APIs), while introducing a rule-based postprocessing module grounded in BPMN 2.0 rules and the block-structured workflow principle to correct the final structure. The TrOCR model achieved a Character Error Rate (CER) of 7.11%, outperforming the reference baseline (8.80%). End-to-end system testing on the test images achieved an Overall F1-Score of 86.0%, with a breakdown of 91.5% Shape F1-Score, 80.6% Edge F1-Score, and 84.8% Label F1-Score. Although per-category performance remained below the reference study, the postprocessing module successfully executed structural corrections with a 100% Fix Rate across eleven test cases, along with a 0% False Positive Rate on images that were already structurally valid without error. A multi-criteria trade-off analysis using the Complex Proportional Assessment (COPRAS) method across four criteria showed that the proposed system achieved higher relative utility than the reference study across all tested weight-sensitivity scenarios.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Block-Structured Workflow, BPMN, BPMN Sketch Recognition, Postprocessing, Transformer OCR. =================================================================== Block-Structured Workflow, BPMN, BPMN Sketch Recognition, Postprocessing, Transformer OCR.
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T11 Technical writing. Scientific Writing T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Naufal Khairul Rizky
Date Deposited:	27 Jul 2026 02:02
Last Modified:	27 Jul 2026 02:02
URI:	http://repository.its.ac.id/id/eprint/137667

Actions (login required)

View Item